DDPG-Based Deep Reinforcement Learning Tactics for Defending Torpedo Attacks

Threats to autonomous underwater vehicle (AUV) operations at sea are numerous, but underwater torpedoes pose the greatest risk since they may be fired in a variety of conditions. Torpedoes are known for their high speed and their ability to utilize acoustic homing towards targets. As AUV lacks rapid evasion speeds, two types of decoys are employed to increase the effectiveness of torpedo countermeasure tactics: mobile decoys and hovering decoys. To effectively counter torpedoes, it is crucial to employ both mobile decoys and hovering decoys in conjunction with AUV maneuvers. In this paper, we propose torpedo countermeasure tactics using a deep deterministic policy gradient (DDPG) algorithm to quickly respond to torpedo threats. This paper designs a reward function based on the distance between the AUV and the torpedos, the survival time of the AUV, and whether or not the AUV is hit. Experiments are conducted by modeling AUV, torpedoes, mobile decoys, and hovering decoys separately, then integrating the defined model. The performance of the model is proven through experiments.